PyOptim Python numerical optimization toolbox

This toolob aim at providing a unified interface to generic optimizers for standard (LP,QP) and gradient based optimization problems (LBFGS, Proximal Splitting, Projected gradient).

As of now it provides the following solvers:

• Linear Program (LP) solver using scipy, cvxopt, or GUROBI solver.
• Quadratic Program (QP) solvers using cvxopt, guroby or quadprog.
• Proximal spliting (a.k.a. ISTA) gradient descent for non smooth optimization.
• Spectral Projected Gradient solvers (spectral is optionnal but strongly recommended).
• Conditional gradient solver.

Planned integration are:

• L-BFGS for smooth optimization (interface to scipy and others)
• Stochastic gradients

Installation

The library has been tested on Linux, MacOSX and Windows.

The following Python modules are necessary and intalled automaticaly with:

• Numpy (>=1.11)
• Scipy (>=1.0)
• Cvxopt

The following modules are optional but obviously necessary to the solvers using them:

• Quadprog
• Gurobipy (official Python interface of GUROBI)
• Stdgrb

You can install the toolbox by downloading it and then running:

python setup.py install --user # for user install (no root)

Contributors

The main contributors of this module are:

• Rémi Flamary
• Stéphane Canu